AI for early outbreak detection

COVID 19 and tuberculosis (TB) have become the new 'cursed duet' of Public health. Both diseases have stressed the health care systems, with COVID 19 pandemic set to worsen the global TB epidemic and reversing years of improvements made. Our AI driven outbreak locating system can predict the COVID 19 cases 2 weeks in advance, while also measuring the risk of TB in the same region. Health authorities can thus strategize in advance, like creating public awareness and preparing for lockdowns, while TB control programs can adapt their services as to minimize interruption of TB diagnosis and treatment. If scaled globally this system can empower the health systems to better prepare for pandemics and prevent drastic blows to the existing national disease control programs like TB, HIV and Malaria. This could help millions of  patients to adhere to their treatments and not suffer in anxiety caused by pandemics and restrictions.     

The root problem we are focusing on is lack of awareness about location and magnitude of disease (TB and COVID 19) and the vulnerable population at risk. As a result, a large number of TB patients go undiagnosed each year, posing hidden risk in the community. Lack of preparedness of communities towards the emergence of COVID 19 caused interruption of health care services, which further aggravated the former. 

Every year around 10 million fall ill with active TB, and 1.4 million lose their lives globally. Three million people with TB are considered missing each year because of under-diagnosis and underreporting. Pakistan has the fifth greatest burden of TB in the world with an estimated 510 000 new TB cases reported every year. Pakistan also reported around 940 000 cases and 21 000 deaths from COVID-19 until now. The country is grappling with over 5000 daily cases and close to 100 deaths every day at present.

COVID 19 lockdowns caused substantial declines in the case notification of drug-susceptible TB by 45% and of drug-resistant TB by 35% in Pakistan. Lack of digitization and smart algorithms lead to a substandard response in controlling spread of diseases. 

The solution is an AI driven disease surveillance platform that is capable of integrating real-time program notification data with remote sensing and other open source data to provide granular subnational disease burden estimates in real time for population clusters of approximately 10 000 people or as defined by local needs. It uses artificial intelligence (AI), data analytics and geographical information systems to quantify risk/burden of infectious diseases like tuberculosis and COVID-19. The outputs can then be visualized on interactive digital dashboards, geoportals and as notifications on mobile phones. Program managers can use the dashboards and geoportals for surveillance and monitoring of the situation. Simpler alerts can be sent on mobile applications and as text messages to community health care workers in resource constrained settings. This integrated TB-COVID 19 surveillance system can give a 14 day forecast on COVID 19 cases in a geographical region, thus enabling health authorities to take preventive measures to control the surge and also adapt services under other disease control programs if lockdowns and major restrictions are anticipated. It can strengthen primary health care systems by providing timely alerts to community health care workers such that continuity of care can be maintained even during pandemics.

Pakistan has a population of 216 million with a population density of 287 per Km2.Pakistan ranked 154th among 189 countries on UN’s Human Development Index (HDI) 2020 rankings. It is a lower middle-income country and the fifth most populous globally. Around 63% of the population lives in rural areas, and 24% of the population lives below the poverty line. Almost half of women have received no education. Over 17 million people do not have access to safe water and more than two in five people don’t have access to sanitation services. The infant mortality rate (IMR) is 74 per 1000 live births, and the maternal mortality ratio (MMR) is 274 per 100,000 live births, higher than that of neighboring countries. Communicable diseases, maternal health and under-nutrition comprise around half of the national burden of disease. It is one of the two remaining countries where Polio is still endemic. 

Pakistan has the 5th highest tuberculosis (TB) burden globally. We have been engaging with the National Tuberculosis Control Program (NTP) since 2019 to help them locate the hotspots of TB and steer their active case finding activities under the electronic case based surveillance (eCBS) project. This project runs in collaboration with an international research institute and local non government organization that plans chest camps and conducts community screening across the region. The project implemented a Tuberculosis eCBS platform that integrates chest camp screening data with contextual variables like population density, environmental features, economic indicators, remote sensing and programmatic data, to provide subnational TB burden estimates for population clusters of approximately 10 000 people, and optimizing the efforts of eCBS program in the region. Our database barely received  new data for almost 8 months in 2020 and 4 months in 2021, communications with the local teams revealed that their screening activities were suspended due to lockdowns, with little clarity on when the activities would be resumed. Published research suggests that between February and April 2020, there were substantial drops in the case notification of drug-susceptible TB by 45% and of drug-resistant TB by 35% in Pakistan. Under the current situation, due to delayed diagnoses and treatment initiation, tuberculosis-related deaths can increase up to 20%  in high burden settings in the next 5 years. The NTP and local teams would like to resume their community screening activities in a phased manner, ensuring the safety of their personnel and a decent yield of cases from the screenings. The current solution can help them achieve the same by predicting COVID 19 cases 2 weeks in advance, and also find the TB hotspots. The local teams which go into the community and the NTP can adapt their movement according to the predicted risk of COVID 19, while safely conducting community case finding for TB. This solution can also support contact tracing for COVID 19 and TB, planning the vaccination policy and monitoring and evaluation of the impact. Both patients on TB treatment and their health care providers can plan in advance and avoid interruption of treatment. 

Problem: inability of healthcare authorities to gauge the magnitude and exact location of risk of infectious diseases like COVID 19 and TB. This magnifies the emergency response mechanism needed to efficiently respond to control spread of diseases, more so in developing countries. 

Solution can:

• Forecast disease risk: can predict the estimated number of COVID 19 cases 2 weeks in advance by looking at existing notifications

• Locate hotspots: Predict risk of disease / estimate number of cases in a given geographical location.

• Improve preparedness: being aware of upcoming waves of COVID surge can soften the blow to the healthcare system 

The company has an existing Tuberculosis electronic case based surveillance program already running  in Pakistan in collaboration with NTP. We also have a validated system running which gives a 14 day forecast on COVID 19 cases in Belgium. The proposed solution is about integrating the two systems together for the first time such that it can predict both diseases in Pakistan. Since the two infectious diseases have considerable similarities in terms of impact on lung function and the sociodemographic profiles of persons at risk, we expect that the individual outcomes from each system can potentiate the other and make the AI algorithm smarter and more accurate. The results from this solution will benefit around 200 million people in Pakistan.

Based on real time Bayesian modeling for integrated TB and COVID 19 risk quantification, the solution combines program notification data with multiple contextual factors like sociodemographic, environmental and remote sensing data associated with diseases. Adaptive feedback mechanism completes the loop of data in real time, such that the system continuously learns from new data coming in from program activities, driven by predicted outcomes. The system becomes smarter and more accurate over time, and remains disease agnostic, so it is scalable to other diseases. The solution also allows real time scenario querying to evaluate the outcome of potential interventions, for example if we get data on X  number of diagnostic tests conducted in the community, what-if queries can help us predict the outcomes and the impact if Y number of tests were to be conducted. Thus it can help to plan the magnitude of response needed to tackle an outbreak in terms of screening, testing and vaccination. The predictions are made on a highly granular level, based on local needs, rather than usual administrative units which can be fairly large and difficult to plan for in times of outbreaks. Leveraging on important contextual risk factors and improved granularity has the potential to storm the market and change the way most disease modelling systems make predictions. Measuring the risk of disease and narrowing down its spatial location in a large administrative unit is the need of the hour and can greatly improve the efficiency of response and its monitoring.

The AI driven system disease risk detection system is being used in all of Pakistan, 4 States in Nigeria nd 2 provinces of South Africa 

Population of Pakistan: 216 M

Population in 4 States of Nigeria: 30 M

Population in South Africa: 10 M

In 1 year = assuming we scale to all of Nigeria and South Africa = 216+ 201 + 58 = 475 Million

In 5 years = assuming we add 2 high TB burden countries like Indonesia and Philippines to the existing countries = 853 Million

EPCON measures the pre-post yield after implementation of the AI-steered system through interactive near real-time dashboards and performance metrics. The predictive performance is constantly measured through retrospective validation, leave one out cross validation and linear regression models.

In addition, user and program meetings provide constant feedback about model output, relevance and user experience.

Some of specific indicators of Public health impact would be:

• Time to detection of outbreaks from the diagnosis of the index case as compared to traditional approach.

• Total yield of community case finding activities, yield of contact tracing activities, case notification rates as compared to historical data

 Company/scalability impact: 

• Extension of the AI driven system towards full regions where applicable (South Africa/Nigeria)

• Leveraging of existing TB detection system towards COVID-19 and measure impact 

• Facilitate COVID-19 country response through partner engagement 

Full time :6

Contractors: 4

The company was formed in 2018 after an initial project on mapping TB hotspots in assignment to the Global Fund. The project involved 10 countries in Southern Africa and taught the founding team about the gaps in the present health care system and especially with regards to TB. Having a strong background in remote sensing and geographical information systems, the team decided to further build on this experience and develop a continuously learning platform that generates a digital representation of the real world from an epidemiological perspective. The recognition and interest in EPCONs technology from Imec in Belgium resulted in a participation and the establishment of EPCON with the sole purpose of developing AI driven predictive models and early outbreak detection systems for public health. 

The founding team has expertise in the medical field, computational biology, machine learning, complex adaptive systems, bayesian theorem and the creation of value driven ecosystems. 

With the core team divided between South Africa, Belgium, Portugal and the Netherlands the team has acquired expertise in bioengineering (2), public health (2), medical (2), machine learning (3), system architect & IoT (2). There are team members based in South Africa and others who have experience working in epidemiology and public health in low middle income settings in Africa and southeast Asia. The combination of varied experiences and cultural diversity help to shape the solutions we design.

As a company active in public health and providing solutions toward low and middle income countries we dedicate a lot of attention to the way we represent our company and position our solutions. Our approach is very much focused on partnerships, co-creation, collaboration and mutual respect. We seek active engagement from the local partners and help them drive the solution towards success. The ultimate goal is to have an in-country partner that can drive the project forward. Our weekly meetings with client country teams allows us to build a rapport with the local stakeholders, learn about their challenges with technology, political pressures and  financial limitations. We engage with clients team members who are directly involved with decision making from all levels of hierarchy and not just the top management.  

With a core team of 10 employees (FTE & contractors) the company employs the following nationalities: South Africa (f=1, m=4) , Portugal (m=1), US (m =1), India (f=1), Belgium (f=1, m=1). 

The company actively works towards a balanced and diverse management and rest of the team, ideally residing throughout various regions of interest.

The MIT Solve platform aligns strongly with EPCONs mission to help fight TB and other health related challenges in low and middle income countries through the use of technology. Furthermore the open and innovation driven approach focussed on co-creation and collaborative environments allow for solutions to be adopted by local stakeholders and thus having an immediate down the line impact. 

MIT Solve global network of innovation partners and access to expert knowledge provides the ideal backing for the type of solutions that EPCON brings to market. Health, business acceleration, machine learning and other domains are still to be explored to full potential within EPCON. 

The additional financial opportunity attached to the challenge enables EPCON to proactively integrate its solutions in Pakistan and help both EPCON and the ecosystem partners prove its value and support the local response towards TB and COVID-19. From a patient's perspective, a responsive government and healthcare provider network will improve the access to care and overall wellbeing.

Business model: we consider EPCON to be immature when it comes down to business model and value based pricing. The target markets space provides significant challenges to establish a sustainable business model with recurring income. 

Towards industry we believe there to be more potential, but here we do not hold the necessary experience to draft a solid licensing structure and business model. 

Monitoring & evaluation: 

Product / Service Distribution (e.g. expanding client base): as we plan on targeting new markets, the product strategy and go to market needs to be adjusted. The engine provides a lot of flexibility and angles at which one can target a specific industry or customer. Knowing how to position the platform so that it creates the needed interest is important.

We would like to partner with other technology based organisations which can complement the services we provide. for example, the organisations that focus on providing diagnostic services in the field of TB like FIND, SystemOne. We could also collaborate with other organizations using AI for health, that build individual level models, optical character recognition system 

The use of large quantities of data in volume and variety that are consumed by EPCON for creating environmental context in combination with health records is at the heart of EPCON scope and expertise. The core engine, making use of supervised and unsupervised Bayesian reasoning to identify patterns within these complex variables and the quantification of risk at the population, community and individual level is quite unique in its kind. Being able to identify which communities are most vulnerable in a country of 200 million plus and enabling the planning of tailored interventions at a neighbourhood resolution provides significant benefits to the National program authorities, as well as the local public and private healthcare providers defining what next steps to take in function of public safety and health. 

From the start (2018) EPCON has been working towards models that can support the TB community in finding the missing cases. The solutions that we have developed since are fully executed according to our initial scope and strategy. Country program data is being consumed by EPCONs engine in combination with contextual data. This creates a real time overview of the situation as is, a digital representation of the real world. Through the engine's learnings from these prior and regional experiences, the platform then propagates the risk of TB across the entire country. The self learning engine continues to learn and improve over time, ensuring an ever evolving platform that helps find the last missing patient. 

Through the agnostic capacity of the engine and the readily available country platforms, it is low hanging fruit to extend the platform towards other disease domains like HIV, Malaria, COVID-19. Being able to visualise the impact of each disease across the nation and having the evidence based tools to drive change is considered to be an opportunity for both the ecosystem partners and the patient community.